JP2023543255A - System and method for detecting gastrointestinal diseases - Google Patents

Abstract

A system includes at least one hardware processor and a non-transitory computer readable storage medium having at least one program code stored thereon. Program code is executable by at least one hardware processor to receive and preprocess the n images. The preprocessing includes at least one of image selection and image adjustment, whereby n' images are acquired and m representations of each of said n' images are created using at least one feature enhancement algorithm. and classifying the n'*m representations into classes by applying a machine learning algorithm to the n'*m representations. The class includes at least positive for gastrointestinal disease and negative for gastrointestinal disease, and when at least a predetermined proportion/percentage of said n'*m expressions are classified as positive for gastrointestinal disease, be identified as suffering from an organic disease. [Selection diagram] Figure 2

Description

The present invention, in some of its embodiments, relates to tongue diagnosis, and more particularly, but not exclusively, to the detection of gastrointestinal disorders.

Tongue diagnosis is a common diagnostic tool in traditional Chinese medicine. Observation of a subject's tongue allows a physician to diagnose the subject's symptoms and/or pathology. Some of the characteristics of the tongue observed by the practitioner are shape, color, texture, geometry, and morphology. By observing such characteristics, a physician can detect a subject's medical condition in a non-invasive manner.

Today, commonly used methods for the detection of lower gastrointestinal pathologies include the fecal guaiac test, the fecal occult blood test (FOBT), and the fecal immunochemical test (FIT).

FIT uses specific antibodies to detect human blood in feces, and it is more conclusive for gastrointestinal pathology than other types of fecal tests, such as the qualitative guaiac fecal occult blood test (FOBT). . The guaiac test often yields false-positive results due to other types of blood that may be present in the digestive system as a result of diet (e.g., red meat) or certain medications. FIT is both more sensitive and more specific than FOBT.

FOBT or FIT generally has a detection sensitivity rate in the range of 40% to 70%. However, it is typically recommended to test subjects three times over three consecutive days using FOBT or FIT to increase the sensitivity of the results. The cost of each kit typically ranges from $7 to $35, and it takes approximately two weeks to receive the results of each laboratory analysis for the test.

Today, a common procedure for detecting upper gastrointestinal pathologies involves gastroscopy, which involves the insertion of a visual aid through an endoscope into the gastrointestinal tract. Therefore, subjects undergo invasive procedures to identify upper gastrointestinal bleeding or other upper gastrointestinal pathologies. Preparation for such treatment includes avoiding food and liquids for 6 to 8 hours prior to treatment.

The foregoing examples of related art and their associated limitations are intended to be illustrative, not exclusive. Other limitations of the related art will become apparent to those skilled in the art from reading this specification and examining the drawings.

The following embodiments and aspects thereof are described and illustrated in conjunction with systems, tools, and methods that are intended to be exemplary and exemplary, and not intended to be limiting in scope.

According to some embodiments of the invention, a system is provided that includes at least one hardware processor and a non-transitory computer-readable storage medium storing program code executable by the at least one hardware processor. Ru. The program code receives n images, each representing the subject's tongue, and preprocesses the n images. The preprocessing includes at least one of image selection and image adjustment, thereby obtaining n' images and using at least one feature enhancement algorithm to create m representations ( presentation). Classify the n'*m representations (i.e., m representations of n' images) by applying a machine learning algorithm to the n'*m representations. The class includes at least positive for gastrointestinal disease and negative for gastrointestinal disease, and when at least a predetermined proportion/percentage of the n'*m expressions are classified as positive for gastrointestinal disease, the subject be identified as suffering from an organic disease.

According to some embodiments of the invention, a computer program product is provided that includes program code executable by at least one hardware processor. The program code receives n images representing the subject's tongue or a portion of the tongue and preprocesses the n images. The preprocessing includes at least one of image selection and image adjustment, thereby acquiring n' images and generating m representations of each of the n' images using at least one feature enhancement algorithm. Then, the n' images are classified into at least two classes by applying a machine learning algorithm to the n'*m representations. When at least two classes include positive for gastrointestinal disease and negative for gastrointestinal disease, and at least a predetermined proportion/percentage of the n'*m expressions are classified as positive for gastrointestinal disease, the subject Identified as having a gastrointestinal disease.

According to some embodiments, image selection includes motion detection. In motion detection, one or more motion vectors are assigned to images captured during motion, and then images whose vectors exceed a predetermined threshold are sorted out from the detected images. According to some embodiments, the image adjustments include adjusting one or more of the n'images' contrast, brightness, level, hue, sharpness, and saturation.
According to some embodiments, the program code determines one or more of a colon-related condition and a stomach-related condition based at least in part on the n'*m expressions that are classified as positive for a gastrointestinal disease. It is possible to further classify the subject into multiple subclasses. According to some embodiments, the subclass further includes two or more subclasses of colon-specific conditions. According to some embodiments, the two or more subclasses of colon-specific pathologies are selected from colorectal cancer (CRC), polyps, different types of polyps, and inflammatory bowel disease including lower intestinal tract (IBD). Ru.

According to some embodiments, the subclass of colon-specific pathology is selected from adenomatous polyps, hyperplastic polyps, serrated polyps, inflammatory polyps, and villous adenomatous polyps, and complex polyps. According to some embodiments, the class further includes two or more subclasses of upper gastrointestinal-specific conditions. According to some embodiments, the two or more subclasses of upper gastrointestinal-specific conditions are selected from gastric malignancies, gastritis, esophageal malignancies, esophagitis, and duodenitis.

According to some embodiments, the subclass includes a score related to the malignancy of the disease. According to some embodiments, a subclass includes a score corresponding to the likelihood that a subject will develop a malignancy in one or more disease states. According to some embodiments, the m representations may further include a three-dimensional representation of the subject's depicted tongue.

According to some embodiments, the program is configured to receive n images from a plurality of different types of imaging devices. According to some embodiments, the program is executable to normalize the received image. According to some embodiments, the hardware processor is coupled to at least one imaging device and the program code is executable to identify the subject's tongue in real time. According to some embodiments, the program code is executable to take n images.

According to some embodiments, a non-transitory computer readable storage medium having at least one hardware data and program code stored thereon is provided. Program code is executable by at least one hardware processor to receive n images. Receive and preprocess n images, each depicting the subject's tongue (or a portion of the tongue). The pre-processing includes at least one of image selection and image adjustment to obtain n' images. The preprocessing involves generating m representations of each of the n' images using at least one feature enhancement and/or extraction algorithm, and applying a machine learning algorithm and optionally a machine learning algorithm to the n'*m representations. The subject is classified based at least in part on the n'*m expressions by applying additional data about the subject. The class includes at least positive for gastrointestinal disease and negative for gastrointestinal disease, and when at least a predetermined proportion/percentage of the n'*m expressions are classified as positive, the subject Identified as having a gastrointestinal disease. According to some embodiments, the additional data about the subject includes, for example, one or more of the subject's blood pressure, heart rate, respiratory rate, age, gender, dietary habits, ethnic background, and smoking habits. may be included.

In some embodiments, the image selection includes motion estimation, where one or more motion vectors are assigned to images taken during the motion, and the vectors are then assigned from the detected images to a predetermined Images that exceed the threshold are excluded. In some embodiments, image adjustments include adjusting one or more of contrast, brightness, texture, levels, hue, and saturation of n' images and/or n'*m representations. include. In some embodiments, image adjustment includes creating additional images from some or all of the n images, such as high dynamic range (HDR) imaging.

In some embodiments, the program code assigns the subject one or more of a colon-related condition and a stomach-related condition based at least in part on the n'*m expressions that are classified as positive for a gastrointestinal condition. Further classification into multiple subclasses is possible. In some embodiments, the subclass further includes two or more classes of colon-specific conditions. In some embodiments, the two or more subclasses of colon-specific pathologies are selected from colorectal cancer (CRC), polyps, and inflammatory bowel disease (IBD). In some embodiments, the subclass further includes two or more classes of gastric-specific conditions. In some embodiments, the two or more subclasses of gastric-specific conditions are selected from gastric malignancies, gastritis, esophageal malignancies, esophagitis, and duodenitis.

In some embodiments, the m representations include three-dimensional representations of the subject's depicted tongue. In some embodiments, the program is configured to receive n images from multiple different types of imaging devices. In some embodiments, the program is executable to normalize the received image. In some embodiments, the m representations include at least one of the original n images (ie, at least one of the n images taken). In some embodiments, the m representations may be images. In some embodiments, the m representations may include files in one or more image formats.

In some embodiments, a hardware processor is coupled to the at least one imaging device and the program code is executable to identify the subject's tongue in real time. In some embodiments, the program code is executable to take n images.

In addition to the exemplary aspects and embodiments described above, further aspects and embodiments will become apparent by reference to the drawings and by consideration of the detailed description below.

Exemplary embodiments are illustrated in the referenced figures. Dimensions of components and features illustrated in the figures are generally chosen for convenience and clarity of presentation and are not necessarily drawn to scale. The drawings are listed below.
FIG. 1 is a schematic simplified diagram of a system for detecting gastrointestinal diseases, according to some embodiments of the invention. FIG. 2 is a flowchart of functional steps in a process for detection of gastrointestinal diseases, according to some embodiments of the invention. FIG. 3 is a front schematic view of an exemplary segment map, according to some embodiments of the invention. FIG. 4 is a simplified perspective view of an exemplary imaging device for detecting gastrointestinal disease, according to some embodiments of the invention.

According to one aspect of some embodiments of the present invention, image processing, computer vision, color science, and/or deep learning is used to determine whether a patient has a gastrointestinal disease based on one or more images of a subject's tongue. Systems and methods are provided for detecting.

In some embodiments, the system includes at least one hardware processor and a storage module with program code stored thereon. In some embodiments, the program code is executable by at least one hardware processor to receive n images, each showing a tongue of a subject, and to preprocess the n images. Pre-processing includes at least one of image selection and image adjustment, thereby obtaining n' images. In some embodiments, the program code generates m representations of each of the n' images using at least one feature enhancement algorithm, and generates m representations of each of the n' images using the at least one feature enhancement algorithm, and generates m representations of each of the n' By applying it to the expressions, it is possible to classify the subject based at least in part on the n'*m expressions. In some embodiments, the class includes at least positive for gastrointestinal disease and negative for gastrointestinal disease. In some embodiments, the program code is configured to identify a subject as having a gastrointestinal disease when at least a predetermined percentage of the n'*m expressions are classified as positive for the gastrointestinal disease. It is doable.

A potential advantage of the systems and methods disclosed herein is that detection of gastrointestinal disease is based on one or more images of a subject's tongue and is therefore non-invasive. In some embodiments, detection of gastrointestinal disease is automatic. In some embodiments, detection is analyzed in real time. In some embodiments, systems and methods for detecting gastrointestinal diseases detect the condition with a sensitivity and/or specificity of at least 70%.

In some embodiments, systems and methods for detecting gastrointestinal diseases include using a common or universal camera to obtain images of the tongue, based on which the pathology is detected. In some embodiments, the systems and methods may be used by a user at home and/or without visiting a practitioner, for example, by taking an image of their tongue using a smartphone camera. . In some embodiments, the system and method for detecting gastrointestinal disease does not require any preparation or special diet during the day or the day before obtaining images of the subject's tongue. Exceptionally, in some embodiments, ingestion of food and/or beverages with pigments/dyes should be prevented prior to obtaining the image. This is to ensure that the coloring of those products does not mask the natural color of the subject's tongue.

A potential advantage of a system and method for the detection of gastrointestinal diseases that does not require any preparation or special diet on or before the day the tongue image is obtained is that the subject does not have to undergo any preparation or special diet leading up to or on the day of the test. There is no need to change your eating habits. For example, oral medications prescribed to a subject can be taken on a regular basis.

In some embodiments, systems and methods for detecting gastrointestinal diseases are configured to replace commonly used detection tests such as FOBT and FIT. In some embodiments, the systems and methods for detecting gastrointestinal disease identify at least three conditions associated with FOBT.

The present disclosure provides one or more machine learning models trained to detect gastrointestinal diseases. It was developed by training a deep neural network using labeled images of the tongues of subjects diagnosed with digestive disorders. In some embodiments, the machine learning models disclosed herein provide higher predictive accuracy compared to known classification techniques. In some embodiments, the present disclosure uses deep learning techniques to generate automated, accurate, standardized machine learning models for early prediction of gastrointestinal diseases.

In some embodiments, the present disclosure provides for training one or more machine learning models based at least in part on training data that includes image data depicting at least a portion of a subject's tongue. . In some embodiments, the image data includes at least one of an image and a series of images. In some embodiments, the image data includes at least one of a video segment, a motion vector segment, and a three-dimensional video segment. In some embodiments, the image data comprises an (original) captured image of the subject's tongue (or part of the tongue). In some embodiments, the image data includes one or more representations of one or more captured images of the subject's tongue. In some embodiments, the representation may be an image. In some embodiments, a representation may include files in one or more image formats.

In some embodiments, the image data depicts the subject's tongue and/or a portion of the tongue. In some embodiments, the image data is captured relative to the time at which the image data was captured and the time at which another image was captured, as described in more detail elsewhere herein. The data includes data indicating at least one of the following relative times. Each possibility is a separate embodiment.

In some embodiments, the image data includes n images for each subject and/or the subject's tongue. In some embodiments, the n images include one or more images. In some embodiments, the n images include multiple images. In some embodiments, image data is acquired using an image sensor such as an active pixel sensor or a charge coupled device (CCD). In some embodiments, image data is acquired using RGB imaging techniques. In some embodiments, image data is obtained using digital topology techniques. In some embodiments, image data may be captured using different imaging equipment. In some embodiments, image data is obtained by focusing all or at least some of the reflected wavelengths from the subject's tongue. In some embodiments, the wavelength ranges from 380 nm to 800 nm. In some embodiments, the reflected wavelength allows for a depth of field of at least 100 mm.

According to some embodiments, image data is acquired under defined illumination of the subject's tongue, as described in more detail elsewhere herein. In some embodiments, image data is obtained while the subject's tongue is illuminated with defined illumination such that the tongue is obtained with optimal color fidelity. In some embodiments, the tongue is illuminated using at least one of a laser and a filter configured to at least partially produce defined illumination. In some embodiments, the system is configured to continuously calibrate the prescribed illumination.

According to some embodiments, the prescribed illumination is configured such that the color of the subject's tongue and the n images taken are metamerically matched. In some embodiments, the system maintains conditional matching between the subject's tongue color and the n images taken, as described in more detail elsewhere herein. The illumination type, illumination spectrum, and/or position of the illumination member is configured to be changed for the purpose of changing the illumination type, illumination spectrum, and/or position of the illumination member.

In some embodiments, a system for detecting gastrointestinal disease is configured to receive, normalize, and/or compare different image data from various imaging devices. In some embodiments, the image data is represented by n' image datasets corresponding to n images. In some embodiments, the n' image datasets are obtained by manipulating the n images in at least one of preprocessing and image processing analysis. In some embodiments, the image manipulation includes at least one of color fidelity, texture enhancement, local contrast enhancement, local color contrast enhancement, geometric feature enhancement, image segmentation, image color segmentation, and motion detection. including. Each possibility is a separate embodiment.

In some embodiments, the manipulation of the images includes a three-dimensional representation of the tongue depicted by the images, motion or motion detection associated with the movement of the tongue during the capture of the images, and n' images and/or n images. generating a video using one or more of the images. Each possibility is a separate embodiment. In some embodiments, the image dataset includes n' manipulated images. In some embodiments, image manipulation includes accounting for motion blur, distortion, and/or data duplication caused by tongue movement during image capture.

In some embodiments, the system and method for detecting a gastrointestinal disease uses a trained machine learning module to detect a gastrointestinal disease based on at least a portion of n' images of an image dataset. configured to detect. In some embodiments, the machine learning module is trained on an image dataset of n' images related to the tongue of one or more subjects with one or more diagnosed gastrointestinal diseases. .

In some embodiments, the machine learning module is configured to receive n images of image data and/or n' images of an image dataset. In some embodiments, the machine learning module is trained to classify received image data into classes and/or subclasses related to gastrointestinal diseases. In some embodiments, a system and/or method for detecting a gastrointestinal disease is configured to generate m representations of each of the n' images. In some embodiments, the system and/or method is configured to generate m representations of at least a portion of the n' images. In some embodiments, the machine learning module is configured to receive m representations. In some embodiments, the machine learning module is trained to classify the m received representations into classes and/or subclasses related to gastrointestinal diseases.

In some embodiments, the class includes at least one of: negative for lower gastrointestinal conditions, positive for lower gastrointestinal conditions, negative for upper gastrointestinal-related conditions, and positive for upper gastrointestinal-related conditions. In some embodiments, the subclass is at least one of colorectal cancer (CRC), polyp, inflammatory bowel disease (IBD), gastric malignancy, gastritis, esophageal malignancy, esophagitis, and duodenitis. including. Each possibility is a separate embodiment.

According to some embodiments, subclasses include types of disease states, such as types of polyps. According to some embodiments, the subclass includes at least one of sessile polyps and pendulum polyps. According to some embodiments, the subclass includes cancerous and/or non-cancerous polyps. According to some embodiments, subclasses include benign and/or malignant polyps. According to some embodiments, the subclass includes one or more of adenomatous polyps, hyperplastic polyps, serrated polyps, inflammatory polyps, and chorioadenomatous polyps, and composite polyps.

According to some embodiments, classifying into subclasses includes ranking the identified diseases with a value associated with the risk level of the disease. According to some embodiments, the level of risk may be the level of lethality and/or urgency of the surgery. According to some embodiments, a subclass includes one or more ranges of polyp sizes.

According to some embodiments, the subclass includes a score related to the level of malignancy of the disease. According to some embodiments, a subclass includes a score corresponding to the likelihood that a subject will develop a malignancy in one or more disease states. For example, according to some embodiments, a subclass may predict the likelihood that a subject without polyps will develop one or more types of polyps. According to some embodiments, the score is rated as a percentage, such as, for example, 0-10%, 10-30%, 30-50%, 50-70%, and greater than 70%. In some embodiments, the score indicates the probability of occurrence of a particular malignancy of a known disease in the subject and/or the probability of occurrence of a particular malignancy of a disease in the subject. According to some embodiments, the score relates to the level of risk that the subject will develop a malignancy in one or more disease states. According to some embodiments, the score is related to the subject's level of risk of developing the condition in the future.

In some embodiments, the system includes at least one hardware processor and a non-transitory computer readable storage medium having program code stored thereon. According to some embodiments, program code is executable by at least one hardware processor to receive and preprocess n images. Each image shows at least a portion of the subject's tongue. The preprocessing includes at least one of image selection and image adjustment, whereby n' images are acquired and m representations of each of the n' images are created using at least one feature enhancement algorithm. The generated n'*m representations are classified into classes using a machine learning algorithm. The class includes at least positive for gastrointestinal disease and negative for gastrointestinal disease, and the subject identify as suffering from a gastrointestinal disease.

According to some embodiments, as described in more detail elsewhere herein, a machine learning algorithm generates data containing n'*m representations related to one or more subjects. Can be trained on set. According to some embodiments, a computer readable storage medium having program code executable by at least one hardware processor is provided to receive and preprocess n images. The program code is executable by at least one hardware processor. Pre-processing includes at least one of image selection and image adjustment. Thereby, acquiring n' images, generating each m representation of the n' images using at least one feature enhancement algorithm, and applying a trained machine learning algorithm to the m representations. is applied to classify the n' images into at least two classes. said at least two classes include positive for gastrointestinal disease and negative for gastrointestinal disease, and when at least a predetermined proportion/percentage of said n' images are classified as positive for gastrointestinal disease, said subject identify as suffering from a gastrointestinal disease.

[Gastrointestinal disease detection system]
Referring to FIG. 1, FIG. 1 is a schematic simplified diagram of a system for detecting gastrointestinal disease, according to some embodiments of the invention. In some embodiments, system 100 includes at least one of a hardware processor 102, a storage module 104, an imaging module 106, an image processing module 108, a machine learning module 110, and a user interface module 112. Each possibility is a separate embodiment. In some embodiments, system 100 is configured to detect a digestive disorder in a subject based on an image of the subject's tongue.

In some embodiments, processor 102 communicates with at least one of storage module 104, imaging module 106, image processing module 108, machine learning model 110, and user interface module 112. In some embodiments, processor 102 is configured to control operation of any one or more of storage module 104, imaging module 106, image processing module 108, machine learning model 110, and user interface module 112. configured. Each possibility is a separate embodiment.

In some embodiments, storage module 104 comprises a non-transitory computer readable storage medium. In some embodiments, storage module 104 comprises one or more program codes and/or sets of instructions for detecting a gastrointestinal disease. Program code directs use of at least one of processor 102, imaging module 106, image processing module 108, machine learning module 110, and user interface module 112. Each possibility is a separate embodiment. In some embodiments, the storage module 104 is configured to detect a digestive disorder in the subject based at least in part on the one or more images of the subject's tongue using the method 200. one or more algorithms.

In some embodiments, imaging module 106 is configured to acquire n images of the subject's tongue. In some embodiments, processor 102 instructs imaging module 106 to acquire one or more of n images. In some embodiments, the imaging module 106 comprises an imaging device and/or a coupler configured to communicate between the system 100 and the imaging device. For example, in some embodiments, imaging module 106 includes a CMOS sensor. In some embodiments, the coupler comprises a cable or wireless connection for processor 102 to acquire n images from the imaging device.

In some embodiments, the imaging module 106 is configured to illuminate the subject's tongue during the capture of the n images. In some embodiments, processor 102 is configured to control illumination of imaging module 106.

In some embodiments, imaging module 106 comprises and/or communicates with one or more sensors configured to detect movement of the subject's tongue. In some embodiments, the one or more sensors are motion sensors, such as thermal sensors.

In some embodiments, one or more of the programs stored in storage module 104 are executable to take n images. In some embodiments, processor 102 is configured to command the capture of images in real time while receiving image data from imaging module 106. In some embodiments, imaging module 106 and/or system 100 are configured to receive images from multiple different types of imaging devices. In some embodiments, system 100 uses image processing module 108 to normalize different images that may be taken by two or more imaging devices, as described in more detail elsewhere herein. configured to do so.

In some embodiments, processor 102 communicates with a cloud storage unit. In some embodiments, storage module 104 comprises a cloud storage unit. In some embodiments, image processing module 108 is stored on cloud storage unit and/or storage module 104. In some embodiments, storage module 104 is configured to receive n images by uploading the n images onto a combined storage unit of storage module 104.
In some embodiments, image processing module 108 is configured to preprocess n images received using imaging module 106. In some embodiments, image processing module 108 is configured to generate an image dataset of n images based at least in part on the n images. In some embodiments, the image processing module is configured to apply an image processing algorithm to at least the distribution of the n images and/or the n' images. In some embodiments, image processing module 108 generates at least a portion of the n' image datasets using an image processing algorithm, as described in more detail elsewhere herein. It is configured as follows.

In some embodiments, machine learning module 110 receives an image dataset of n'*m representations. In some embodiments, as described in more detail elsewhere herein, machine learning module 110 determines one or more digestive organs associated with an image dataset of n'*m representations. Trained to detect diseases.

In some embodiments, system 100 includes a user interface module 112. In some embodiments, the user interface module 112 includes metadata such as, for example, age, gender, blood pressure, dietary habits, risk factors associated with a particular disease, genetic data, the subject's family medical history, and the subject's medical history. is configured to receive from the user. In some embodiments, user interface module 112 communicates with the processor such that user input data is provided to machine learning module 110.

In some embodiments, user interface module 112 includes at least one of a display screen and buttons. In some embodiments, user interface module 112 comprises software configured to transfer input information from a user to processor 102. In some embodiments, user interface module 112 comprises a computer program and/or a smartphone application.

In some embodiments, user interface module 112 includes a keyboard. In some embodiments, a user interface module is configured to receive data from processor 102 and/or display data received from processor 102. In some embodiments, user interface module 112 is configured to display results of gastrointestinal disease detection.

[Method for detecting gastrointestinal diseases]
FIG. 2 is a flowchart of functional steps in a process for detecting gastrointestinal disease, according to some embodiments of the invention.

In some embodiments, method 200 includes acquiring one or more images of the subject's oral cavity. In some embodiments, method 200 includes acquiring n images of the subject's tongue. In some embodiments, the n images include at least one image. In some embodiments, the n images include a number of images ranging from 1 to 10. In some embodiments, the n images include a number ranging from 6 to 45 images. In some embodiments, method 200 includes taking n images of the subject's tongue. In some embodiments, method 200 includes receiving n images directly and/or indirectly from an imaging device. In some embodiments, method 200 includes communicating with imaging device and/or imaging module 106.

In some embodiments, method 200 includes acquiring one or more images of a subject's tongue with a particular band of illumination. In some embodiments, the band of illumination includes multiple wavelengths. In some embodiments, the band of illumination includes wavelengths in the range of about 380-730 nm, such as about 400 nm-700 nm.

In some embodiments, at step 202, method 200 includes receiving n images, each showing the subject's tongue. In some embodiments, a method includes acquiring n images via a connector coupled to an imaging device. In some embodiments, method 200 includes receiving n images from imaging module 106 and/or storage module 104. In some embodiments, the n images are transferred from the imaging module 106 to one or more of the processor 102, the storage module 104, the image processing module 108, the machine learning model 110, and the user interface module 112. Ru.

In some embodiments, method 200 includes identifying the subject's tongue and/or the position of the subject's tongue in at least one of the n images. In some embodiments, method 200 includes identifying the subject's tongue and/or the position of the subject's tongue in real time. In some embodiments, method 200 includes identifying tongue position using one or more motion sensors.
In some embodiments, method 200 includes capturing an image of the subject's tongue in a predetermined location.

In some embodiments, method 200 includes capturing an image of the subject's tongue at specified lighting settings. In some embodiments, the method instructs the imaging module 106 to take images at specified times and/or in real time corresponding to one or more identified locations of the subject's tongue. including doing.

In some embodiments, processor 102 is configured to identify the predetermined location of the tongue. In some embodiments, processor 102 is configured to identify specified lighting settings. In some embodiments, processor 102 is configured to identify a position on the tongue where the back of the tongue is exposed. In some embodiments, the predetermined location is a tongue location in which any one or more of the tongue apex, tongue body, midlingual groove, valve papillae, and lobular papillae of the tongue are exposed. including. In some embodiments, the predetermined position of the tongue includes a position where the subject's tongue extends outward. In some embodiments, the predetermined position of the tongue includes a position where at least a portion of the dorsal surface of the tongue is parallel to the imaging device.

According to some embodiments, the predetermined position of the tongue is associated with visually exposing a segment of the tongue according to a particular tongue reflex segment of the tongue. In some embodiments, method 200 includes identifying one or more tongue reflex segments of the subject's tongue.

In some embodiments, method 200 includes applying a calibration process configured to capture images of the tongue. In some embodiments, one or more of the calibration processes includes image segmentation, real-time image calibration, avoiding lighting clipping, avoiding lighting clipping in real-time, and avoiding motion blur. and avoiding motion blur in real time. Each possibility is a separate embodiment.

In some embodiments, method 200 includes verifying uniformity of illumination of the images during and/or after taking the one or more n images. In some embodiments, method 200 includes verifying the focus level of the images during and/or after taking the one or more n images. In some embodiments, method 200 includes controlling shading of a lens of an imaging device. In some embodiments, method 200 includes calibrating images taken by an imaging device. In some embodiments, method 200 includes calibrating n images.

In some embodiments, method 200 includes illuminating the subject's tongue with defined illumination. In some embodiments, method 200 includes illuminating the subject's tongue such that the subject's tongue is obtained with optimal color fidelity. In some embodiments, method 200 includes calibrating the defined illumination prior to acquiring image data. In some embodiments, method 200 includes continuously calibrating the prescribed illumination. In some embodiments, the method 200 determines the illumination type, illumination spectrum and and/or adjusting the position.

In some embodiments, method 200 includes capturing a video and/or a plurality of images. In some embodiments, method 200 includes uploading captured images, images, and/or videos to a cloud storage unit. In some embodiments, method 200 includes analyzing tongue movement vectors depicted by a video and/or a plurality of images. In some embodiments, method 200 includes tracking the tongue. In some embodiments, method 200 includes tracking the tongue in real time and/or in captured video segments. In some embodiments, method 200 includes analyzing tongue vibrations. In some embodiments, method 200 includes analyzing tongue movement vectors. In some embodiments, method 200 includes analyzing motion associated with motion of the imaging device. In some embodiments, a method includes distinguishing between motion associated with tongue vibration and motion associated with imaging device motion.

A potential advantage of analyzing tongue movements is that the analysis allows for the removal of blur caused by tongue vibrations, allowing for the generation of clearer images.

In some embodiments, method 200 includes motion estimation, and one or more motion vectors are assigned to images captured during motion. In some embodiments, the method includes excluding detected images whose vectors exceed a predetermined threshold. In some embodiments, the predetermined threshold corresponds to a predetermined resolution of the processed image.

In some embodiments, the method 200 includes obtaining n' images by preprocessing the n images in step 204. In some embodiments, pre-processing includes at least one of image selection and image adjustment. In some embodiments, preprocessing includes normalization of the n images. In some embodiments, the image adjustments include adjusting one or more of n images' contrast, brightness, level, hue, and saturation. Each possibility is a separate embodiment. In some embodiments, a method includes preprocessing at least a portion of the n images.

According to some embodiments, pre-processing includes generating segmentations that represent specific regions of the tongue within the n images. According to some embodiments, one or more of the designated regions of the tongue may be associated with one or more reconnections of the gastrointestinal tract. According to some embodiments, one or more of the designated regions of the tongue may correspond to one or more pathologies of the gastrointestinal tract. According to some embodiments, each specific region of the tongue corresponds to a respective pathology of the gastrointestinal tract.

Reference is made to FIG. 3, which is a front schematic illustration of an exemplary segment map, according to some embodiments of the invention. According to some embodiments, segments 302a/302b/302c/302d/302e/302f/302g/302h/302i/302j/302k/302l/302m (hereinafter collectively referred to as segments 302) are located on the subject's tongue. A map 300 of at least a portion of the surface is defined. According to some embodiments, segment 302 defines at least 2, 3, 5, 8, or 10 designated regions. According to some embodiments, two or more segments 302 are adjacent. According to some embodiments, two or more segments 302 are congruent. According to some embodiments, two or more segments 302 coincide.

According to some embodiments, preprocessing includes assigning a key to each of the segments of segment 302, such as, for example, a numerical value and/or a color. According to some embodiments, the key indicates a particular region of the tongue, eg, the tip, base of the tongue. According to some embodiments, map 300 of segments 302 may be represented by keys associated with segments 302. According to some embodiments, pre-processing includes mapping the subject's tongue based at least in part on the segments 302 and the associated keys of each of the segments.

According to some embodiments, method 200 includes inputting segment 302 and a key associated with segment 302 to a machine learning module. According to some embodiments, method 200 includes assigning an index value to the input key, where the index value is associated with a particular medical condition. According to some embodiments, method 200 includes inputting the indicator value into a machine learning module. According to some embodiments, the index value of the key is associated with the relevance of a particular medical condition to the specified region of the tongue associated with the key. According to some embodiments, the index value is binary. According to some embodiments, the index value includes a range that corresponds to the level of association between a particular region and a particular disease state.

For example, to detect a particular disease state or its risk of malignancy, there may only be one or two relevant segments 302, such as segments 302a and 302j. In such an example, the index values assigned to other segments such as 302d and 302k may be "0". The index value assigned to 302a and 302j may be "1".

According to some embodiments, the machine learning module is configured to analyze the n images according to the index values assigned to the keys and/or segments 302. According to some embodiments, the machine learning module is configured to overlook segments where the index value associated with the key corresponds to a low level of relevance for a particular condition. For example, in some embodiments, the machine learning module is configured to overlook segments where the index value associated with the key is "0".

In some embodiments, method 200 includes evaluating the quality of n images. In some embodiments, processor 102 receives quality data for n images. In some embodiments, method 200 includes directing changes to imaging module 106 based on the evaluation. In some embodiments, for images where a change in ambient conditions during image capture may improve the resulting quality of the image, processor 102 directs the imaging module 106 to make changes such that the change in ambient conditions occurs. be able to. For example, the processor can command changes in the brightness around the subject while taking images. In some embodiments, method 200 includes instructing a user to make position and/or lighting changes using user interface module 112. In some embodiments, processor 102 is configured to directly command changes to the imaging device during the capture of the n images, as described in more detail elsewhere herein.

In some embodiments, the method 200 includes determining the angle of the imaging device, the light emitted from the illumination member of the imaging device, the brightness of the illumination of the subject during imaging, the angle of the imaging device relative to the tongue of the subject, and during imaging. including adjusting any one or more of the exposure times used. In some embodiments, the method includes determining any one or more of the following: imaging module and/or camera settings, background, background, magnification, and magnification of a particular region of a subject's tongue and/or oral cavity. including adjusting the

In some embodiments, method 200 includes combining two or more of the n images to form a single image of n' images. In some embodiments, a method includes combining a plurality of n images to form a plurality of n' images. In some embodiments, the method includes stitching images. In some embodiments, a method includes stitching n images to obtain n' images. In some embodiments, a method includes stitching n images together. In some embodiments, the method stitches the n images such that the single stitched image each includes data for a different angle of the tongue depicted by a separate image of the n images. Including matching. In some embodiments, a method includes generating n' images from the n images, such that the n' images include super-resolution images. In some embodiments, the number of n' images corresponding to the n images is less than the number of n images acquired. In some embodiments, the number of n' images corresponding to the n images is equal to the number of n images acquired.

In some embodiments, the number of n' images corresponding to the n images is greater than the number of n images acquired. In some embodiments, the n' images include data from one or more of the n images, eg, indicative of particular features of the n images. In some embodiments, the n' images include partial data from one or more of the n images. In some embodiments, one or more of the n' images includes partial data from multiple n images.

In some embodiments, method 200 includes obtaining high dynamic range (HDR) imaging using n images. In some embodiments, method 200 includes image conditioning that includes creating one or more images from at least a portion of the n images. According to some embodiments, the n images are acquired using one or more different focus or exposure levels. According to some embodiments, at least a portion of the n images are composited into one or more images, thereby providing optimal exposure for each portion of the subject's tongue within the one or more images. obtain.

In some embodiments, the number of n' images corresponding to the n images is less than the number of n images acquired due to the sorting of the n images. In some embodiments, one or more of the n images have one of the following: resolution below a predetermined threshold, contrast below a predetermined threshold, and contrast above a predetermined threshold. is not associated with n' images for one or more reasons. In some embodiments, processor 102 has determined the predetermined threshold. In some embodiments, a user can determine the predetermined threshold via user interface module 112.

In some embodiments, method 200 includes applying an image processing algorithm to n images. In some embodiments, method 200 includes generating m representations of each of the n' images in step 206. In some embodiments, method 200 includes generating m representations based on n' images. In some embodiments, method 200 includes generating m representations of each of the n' images using at least one feature enhancement and/or extraction algorithm.

In some embodiments, one or more of the m representations include one or more features extracted from the n images. In some embodiments, one or more of the m representations comprises one or more of the n images. In some embodiments, the m representations may be images. In some embodiments, the m representations may include files in one or more image formats.

In some embodiments, one or more of the m representations may include a tongue segment color map, a motion vector associated with the n images, a two-dimensional representation of the subject's tongue, or a three-dimensional representation of the subject's tongue. a representation, a representation of different planes of the tongue, multiple positions of the tongue, a topology map of the tongue, and one or more digitally generated video segments. In some embodiments, the m representations include positions and/or configurations of the subject's tongue that are digitally generated without images of the tongue at the particular positions represented in the representations. In some embodiments, the m representations include functions of time.

In some embodiments, one or more of the m representations include one or more features extracted from the n images. In some embodiments, the extracted features include color, morphology of one or more of the tongue surfaces, tongue topology, tongue dimensions, and tongue vibration analysis. Each possibility is a separate embodiment.

In some embodiments, at step 208, method 200 includes classifying the n'*m expressions into classes. The class includes at least positive for lower gastrointestinal pathology and negative for lower gastrointestinal pathology. According to some embodiments, classifying the n'*m representations includes applying the trained machine learning module 110 to the n'*m representations. In some embodiments, the class includes at least positive for gastrointestinal disease and negative for gastrointestinal disease. In some embodiments, the method includes individually classifying the subject based on each of the n' images. In some embodiments, a method includes classifying a subject based on each of the m expressions.

In some embodiments, method 200 includes further classifying subjects classified as positive for a gastrointestinal disease into one or more subclasses of colon-related conditions and stomach-related conditions. In some embodiments, method 200 subclassifies a colon-related condition into one or more of the following: colorectal cancer (CRC), polyps, and inflammatory bowel disease (IBD), including lower intestinal tract (colon). Including. In some embodiments, method 200 includes subclassing the upper gastrointestinal condition into one or more of gastric malignancy, gastritis, esophageal malignancy, esophagitis, and duodenitis.

In some embodiments, at step 210, method 200 includes identifying the subject as suffering from a gastrointestinal disease. In some embodiments, the method determines whether a subject has a gastrointestinal disease based, at least in part, on the percentage of n'*m expressions associated with the subject being positive for the one or more gastrointestinal diseases. including identifying that In some embodiments, the method determines whether the subject belongs to a subclass of gastrointestinal disease based, at least in part, on the percentage of n'*m expressions that are associated with the subject being positive for a particular subclass. Including identifying as having the disease. In some embodiments, a method includes identifying a subject as suffering from a gastrointestinal disease based at least in part on the proportion of n'*m expressions associated with each of the subclasses.

In some embodiments, method 200 includes presenting the detected condition to a user using user interface module 112.

[Machine learning module and training]
In some embodiments, machine learning module 110 determines one or more digestive disorders based on one or more images of the tongue of a subject identified as suffering from one or more digestive disorders. Equipped with a machine learning module trained to detect diseases. In some embodiments, the machine learning model is based at least in part on metadata related to the subject, such as, for example, one or more of the subject's age, gender, blood pressure, dietary habits, and medical history. trained to detect one or more gastrointestinal diseases. In some embodiments, the sensitivity of detection is greater than at least one of 70%, 80%, 90%, 95%, and 98%. Each possibility is a separate embodiment.

In some embodiments, the machine learning model uses all n images and/or m representations to detect gastrointestinal disease. In some embodiments, a machine learning module uses all of the metadata to detect gastrointestinal diseases. In some embodiments, the machine learning module is configured to use all of the received data to detect one or more gastrointestinal diseases and/or to identify a medical condition associated with the subject. at least one neural network configured to.

In some embodiments, a machine learning model is trained on a training set that includes m representations associated with n images of a subject with an identified gastrointestinal disease. In some embodiments, a machine learning model is trained on a training set that includes m representations associated with n images of subjects identified as not suffering from a gastrointestinal disease. In some embodiments, a machine learning model is trained on a training set that includes m representations associated with n images of a subject identified as having a gastrointestinal disease. In some embodiments, a machine learning model is trained on a training set that includes m representations associated with n images of subjects identified as not suffering from a gastrointestinal disease. In some embodiments, a machine learning model is trained on a training set that includes n'*m representations of subjects identified as having a gastrointestinal disease. In some embodiments, the m representations of the training set include at least an indicator indicating positive for a gastrointestinal disease and negative for a gastrointestinal disease. In some embodiments, the label is associated with a colon-related condition, a lower GI-related condition, an upper GI-related condition, an upper GI-related condition, colorectal cancer (CRC), polyp, inflammatory bowel disease (IBD), Indicates any one or more of gastric malignancy, gastritis, esophageal malignancy, esophagitis, and duodenitis.

In some embodiments, the same machine learning model is trained for multiple gastrointestinal diseases. In some embodiments, the same machine learning model is trained using labels associated with multiple gastrointestinal diseases.

In some embodiments, the labels are associated with different segments of the tongue indicated by n'*m representations. In some embodiments, a single image is associated with multiple labels, each label being associated with a different segment of the tongue depicted by the image. In some embodiments, the labels relate to tongue segments indicated by n'*m expressions. In some embodiments, a single image is associated with multiple labels, each label associated with a tongue reflex segment of the subject's tongue that is depicted.

[Imaging device]
Reference is made to FIG. 4 illustrating a simplified perspective view of an exemplary imaging device for detecting gastrointestinal disease, according to some embodiments of the present invention. In some embodiments, the imaging device 400 is configured to allow uniform positioning of the subject's tongue relative to the camera and/or lens of the imaging device 400. In some embodiments, imaging device 400 is in operative communication with at least one of imaging module 106 and processor 102. Each possibility is a separate embodiment. In some embodiments, imaging device 400 is configured to allow a user to capture images of a subject's tongue.

In some embodiments, imaging device 400 includes frame 402. In some embodiments, frame 402 includes a base 404 and at least one leg 406. In some embodiments, frame 402 is rigid. In some embodiments, base 406 and/or legs 406 are configured to stabilize frame 402 when positioned on a surface. In some embodiments, frame 402 includes two or more struts 416 extending from one or more of base 404 and legs 406. In some embodiments, two or more posts 416 are configured to support at least one of rest 410 and holder 408. Each possibility is a separate embodiment.

In some embodiments, rest 410 includes a forehead rest and is configured to maintain a stationary position of the subject's head relative to frame 402. In some embodiments, rest 410 is configured to abut a portion of the subject's face. In some embodiments, the rest 410 is coupled at a first end to one post 416 and at a second end to a second post 406. In some embodiments, rest 410 is rigid, semi-rigid, or flexible. In some embodiments, rest 410 is adaptable so that the subject's particular facial features may be taken into account during use. In some embodiments, the position of rest 410 is adjustable with respect to frame 402 and/or at least one post 416.

In some embodiments, the imaging device 400 includes a holder 408 configured to fix the position of the camera and/or sensor. In some embodiments, imaging device 400 includes a camera and/or a sensor. In some embodiments, holder 408 includes a dock onto which a camera and/or sensor is coupled. In some embodiments, the dock is configured to receive a general purpose camera, such as a smartphone, for example. In some embodiments, the dock is slidably coupled to frame 402. In some embodiments, the holder 408 and/or the dock are slidable about fewer or at least two axes of movement. In some embodiments, the holder 408 and/or the dock are slidable about at least three axes of movement. In some embodiments, the dock is tiltable such that the spatial orientation of cameras and/or sensors coupled to holder 408 is changed. In some embodiments, the dock is adjustable such that the camera and/or sensor coupled to holder 408 is tilted, panned, and/or rolled. In some embodiments, the camera is configured to pan, tilt, and/or roll relative to any one or more of frame 402, rest 410, and holder 408. In some embodiments, the angle of the dock relative to frame 402 and/or rest 410 is adjustable. In some embodiments, holder 408 includes a motor configured to drive holder 408 into a predetermined position and/or angle relative to frame 402.

In some embodiments, imaging device 400 includes a processor 418. In some embodiments, imaging device 400 is in operative communication with processor 102/418. In some embodiments, the imaging device 400 includes a power supply unit 414 in communication with the processor 102/418. In some embodiments, a power unit 414 is coupled to the motor of holder 408.

In some embodiments, processor 102/400 is configured to directly command changes in the structure of imaging device 400 during the capture of n images. In some embodiments, processor 102/400 is configured to directly command changes in the position and/or angle of holder 408 relative to frame 402.

Unless otherwise defined, various embodiments of the present invention may be provided to end users in multiple formats, platforms, computer readable memory, computer display devices, printouts, networked computers, tablets, etc. The information may be output to at least one of a smartphone application or a user. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The materials, methods, and examples provided herein are illustrative only and not intended to be limiting.

Implementation of the methods and systems of the present invention includes performing or completing certain selected tasks or steps manually, automatically, or a combination thereof. Furthermore, according to the actual implementation and equipment of the preferred embodiments of the method and system of the present invention, some selected steps can be performed by hardware or by software on any operating system in any firmware. or a combination thereof. For example, as hardware, selected steps of the invention may be implemented as a chip or a circuit. As software, selected steps of the invention may be implemented as software instructions executed by a computer using any suitable operating system. In either case, selected steps of the methods and systems of the present invention may be described as being performed by a data processor, such as a computing platform, for executing instructions.

Although the present invention is described in terms of a "processor" or "computer" on a "computer network," optionally any device that features a data processor and/or limited ability to execute one or more instructions. Note that computers may be described as computers, including, but not limited to, PCs (personal computers), servers, minicomputers, mobile phones, smart phones, PDAs (personal digital assistants), and pagers. Two or more such devices in communication with each other and/or any computer in communication with any other computer may optionally include a "computer network."

Embodiments of the invention may include apparatus for performing the operations herein. The apparatus may be specially configured for the desired purpose or may include a general purpose computer selectively activated or reconfigured by a computer program stored on the computer. Such computer programs can be stored on floppy disks, optical disks, CD-ROMs, magneto-optical disks, read-only memory (ROM), random access memory (RAM), electrically programmable read-only memory (EPROM), electrical including, but not limited to, erasable and programmable read-only memory (EEPROM), magnetic or optical cards, or any other type of medium suitable for storing electronic instructions and that can be coupled to a computer system bus. , may be stored on any type of computer-readable storage medium, such as a disk.

The processes and displays presented herein are not inherently related to any particular computer or other apparatus. Various general purpose systems may be used with programs in accordance with the teachings herein, or it may prove convenient to construct more specialized apparatus to carry out the desired method. The desired structure for a variety of these systems will appear from the description below. Furthermore, embodiments of the invention have not been described with reference to any particular programming language. It will be appreciated that a variety of programming languages may be used to implement the inventive teachings described herein.

The invention may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The invention may also be practiced in distributed computing environments where tasks are performed by distributed computing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.

Although several exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain modifications, additions, and combinations thereof. It is therefore intended that the following claims be construed as including all modifications, additions, and subcombinations as fall within their true spirit and scope.

Throughout this application, various embodiments of the invention may be presented in a range format. It is to be understood that the description in range format is merely for convenience and brevity and is not to be construed as an inflexible limitation on the scope of the invention. Accordingly, range statements should be considered to include all possible subranges specifically disclosed, as well as individual numerical values within that range. For example, the description of a range such as 1-6 includes specifically disclosed subranges such as 1-3, 1-4, 1-5, 2-4, 2-6, 3-6, etc., as well as subranges within that range. It should be considered to have individual numbers, for example 1, 2, 3, 4, 5, and 6. This applies regardless of the scope.

Whenever a numerical range is indicated herein, it is meant to include any recited numbers (fractional or integer) within the indicated range. The phrases "range between" a first index and a second index and the phrases "range between" a first index and "to" a second index are used interchangeably herein. , is meant to include the first index and the second index, and include all fractions and integers therebetween.

In the description and claims of this application, each of the words "comprising," "comprising," and "having" and their formations are not necessarily limited to the members in the list with which the words may be associated. Additionally, in the event of any discrepancy between this application and any document incorporated by reference, it is hereby intended that the present application shall control.

The descriptions of various embodiments of the invention are presented for purposes of illustration and are not intended to be exhaustive or limited to the disclosed embodiments. It will be apparent to those skilled in the art that many modifications and variations are possible without departing from the scope and spirit of the described embodiments. The terminology used herein is used to best describe the principles of the embodiments, their practical application or technical improvements to technology found in the marketplace, or to help those skilled in the art understand the embodiments disclosed herein. selected to enable understanding.

Claims (17)

at least one hardware processor;
a non-transitory computer-readable storage medium storing program code executable by the at least one hardware processor, the system comprising:
The program code causes the at least one hardware processor to:
receiving n images, each image showing at least a portion of the subject's tongue;
preprocessing the n images, the preprocessing including at least one of image selection and image adjustment, thereby obtaining n'images;
generating m representations of each of the n′ images using at least one feature enhancement algorithm;
Classifying the generated n'*m expressions into classes by applying a machine learning algorithm to the n'*m expressions;
said class comprises at least positive for gastrointestinal diseases and negative for gastrointestinal diseases, if at least a predetermined proportion/percentage of the n'*m expressions generated are classified as positive for gastrointestinal diseases; A system in which a subject is identified as suffering from a gastrointestinal disease. 2. The image selection method according to claim 1, wherein the image selection includes motion estimation, wherein one or more motion vectors are assigned to images captured during motion, and wherein the motion vectors exclude images that exceed a predetermined threshold. System described. According to any one of claims 1 to 2, the image adjustment includes adjusting one or more of contrast, brightness, level, hue, sharpness, and saturation of the n' images. system. The program code assigns the subject to one or more subclasses of a colon-related condition and a stomach-related condition based at least in part on the n'*m expressions being classified as positive for a gastrointestinal disease. A system according to any one of claims 1 to 3, which is executable for further classifying. 5. The system of claim 4, wherein the classification into subclasses further includes two or more subclasses of colon-specific conditions. 6. The two or more subclasses of colon-specific pathologies are selected from colorectal cancer (CRC), polyps, different types of polyps, and inflammatory bowel disease including lower intestinal tract (IBD). system. 7. The system of claim 6, wherein the subclass of colon-specific pathologies is selected from adenomatous polyps, hyperplastic polyps, serrated polyps, inflammatory polyps, and chorioadenomatous polyps, and complex polyps. The system according to any one of claims 5 to 7, wherein the classification into subclasses further includes two or more subclasses of upper gastrointestinal specific pathologies. 9. The system of claim 8, wherein the two or more subclasses of upper gastrointestinal specific conditions are selected from gastric malignancy, gastritis, esophageal malignancy, esophagitis, and duodenitis. A system according to any one of claims 5 to 9, wherein the subclass comprises a score related to the level of malignancy of the disease. The system according to any one of claims 5 to 10, wherein the subclass comprises a score corresponding to the likelihood that the subject will develop a malignancy in one or more disease states. The system according to any one of claims 1 to 11, wherein the m representations may further include a three-dimensional representation of the subject's depicted tongue. A system according to any preceding claim, wherein the program is configured to receive the n images from a plurality of different types of imaging devices. 14. The system of claim 13, wherein the program is executable to normalize the received image. 15. The system of any one of claims 1-14, wherein the hardware processor is coupled to at least one imaging device and the program code is executable to identify a subject's tongue in real time. . 16. The system of claim 15, wherein the program code is executable to capture the n images. A computer program product comprising a computer readable storage medium embodying program code executable by at least one hardware processor, the computer program product comprising:
The program code causes the at least one hardware processor to:
receiving n images, each image showing at least a portion of the subject's tongue;
pre-processing the n images, the pre-processing comprising at least one of image selection and image adjustment, thereby obtaining n'images;
generating m representations of each of the n′ images using at least one feature enhancement algorithm;
Classifying the generated n'*m representations into at least two classes by applying a trained machine learning algorithm to the n'*m representations;
the at least two classes include positive for gastrointestinal disease and negative for gastrointestinal disease;
A computer program product, wherein a subject is identified as suffering from a gastrointestinal disease when at least a predetermined proportion/percentage of the n'*m representations generated are classified as positive for the gastrointestinal disease.

Images